0
$\begingroup$

I have implemented beamformer(MVDR/LCMV) for speech enhancement in matlab, now I am trying to convert it to fixed point implementation. during the weight calculation step correlation matrix inversion is required, but in fixed point implementation correlation matrix inversion is not giving correct results. I tried the the standard method math.stackexchange.com/questions/623780/… I am using Q16 format for this(I think this will not work at all as precision will be lost in devision). I have also tried DCD algorithm explained in etheses.whiterose.ac.uk/1629/1/PhD_thesis_Jie_Liu.pdf but not getting the correct result.

$\endgroup$
  • $\begingroup$ You need to give us more information. What algorithm are you using to calculate the matrix inverse now? Have you used all the standard tricks for fixed point implementation? In any event, your question probably better resides on SO itself. See, for example, this question. $\endgroup$ – Peter K. May 21 '15 at 16:48
  • $\begingroup$ I tried the the standard method math.stackexchange.com/questions/623780/… I am using Q16 format for this(I think this will not work at all as precision will be lost in devision). I have also tried DCD algorithm explained in etheses.whiterose.ac.uk/1629/1/PhD_thesis_Jie_Liu.pdf but not getting the correct result. $\endgroup$ – arpit jain May 22 '15 at 5:35
  • $\begingroup$ As Peter says, please provide a bit more information (possibly the equation you are trying to implement). And also, how are you trying to calculate the matrix inverse? Is this going to be done in real time? If so, inverting a matrix may be to costly for real time, you may want to try other adaptive methods. $\endgroup$ – bone Nov 17 '15 at 8:55

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.